Summer is not exactly slowing down.
Washington is expanding the boundaries of its technology competition with China from tariffs and defense procurement into robots, power infrastructure and potentially the optical components sitting inside AI data centers. China is responding. Meanwhile, AI companies are confronting a different problem: increasingly capable agents are finding vulnerabilities, companies are experimenting with how to deploy them at work, and the labor market around AI is beginning to look very different from the one Silicon Valley spent the last few years predicting.
And because not everything on the radar needs to be a national-security crisis, we end with British folklore, old stories and why people seem increasingly interested in looking backward while technology accelerates forward.
On My Radar is dispatch series on what I’m paying attention to right now and the underlying questions guiding Potentia: who actually holds power, and how do you build your own?
Here’s what’s on my radar.
Tune in and listen or watch! We hope you enjoy it as much as we did.
“AI can help with structure and framing. It can’t replace discernment or decision-making.” — Chelsea Toczauer
Washington Keeps Expanding the China Technology Perimeter
Before summer, the Pentagon placed Alibaba and BYD on a defense-related blacklist, prompting lobbying firms to drop the companies before Alibaba challenged the government in court. Then came another round of Section 301 tariffs covering dozens of economies. In July, the FCC moved against foreign-built humanoid and quadruped robots and connected power inverters, a category where Chinese manufacturing is particularly important.
The common thread is that Washington is increasingly treating technological dependence itself as a security vulnerability. That logic may now reach an obscure but critical piece of AI infrastructure: optical transceivers.
These devices convert electrical signals into light and back again, allowing massive amounts of information to move over fiber-optic networks. Inside AI data centers, they connect the thousands of chips working together to train and run increasingly large models. They are not flashy. They are also exactly the kind of component that becomes impossible to ignore once governments begin asking what happens if a geopolitical competitor controls too much of the supply chain.
The Trump administration’s argument is essentially preventive. China has already demonstrated its willingness to use control over strategic materials such as rare earths during trade disputes. Washington does not want to discover during a future crisis that a similar dependency exists inside the infrastructure supporting American AI. But there is a catch: removing a dependency before alternative supply chains exist might create a new vulnerability of its own.
China, unsurprisingly, is responding to the onslaught of US actions. Beijing announced countermeasures including restrictions involving drones and dealings with several American entities, while broader defense contacts between the two governments also appear strained.
Data Centers Are Becoming Political Infrastructure
At the same time Washington is worrying about what goes inside AI data centers, American communities are increasingly arguing about whether they want the facilities at all.
States from New York to even Texas has enacted the growing debate around data-center moratoriums, pauses and restrictions. Meanwhile, development is increasingly looking westward toward areas where land is cheaper and energy, whether renewable or fossil fuel, is easier to pair with enormous new computing loads.
The United States is simultaneously encouraging extraordinarily rapid AI adoption while grappling with concerns about job displacement and relatively limited protections for workers whose roles may change. China is experimenting with a different balance between automation and social protection. Neither approach has resolved the underlying problem: companies have economic incentives to automate faster than institutions can redesign work around the technology.
AI Agent Hacks
The recent in a series of incidents involving increasingly autonomous AI systems identifying vulnerabilities and conducting activity that would previously have required much more direct human involvement.
Hugging Face — its account of the July incident
Hugging Face — technical timeline
Anthropic — investigating real-world incidents during cyber evaluations
OpenAI — subsequent third-party cyber evaluations
OpenAI has characterized this broader development as an important threshold in autonomous cyber capability, while similar experiments are now appearing across several frontier AI companies. The risk for AI companies is quickly becoming less theoretical.
AI Labor Market Trends
The jobs emerging around this transition are also not necessarily the ones people expected. Companies and governments are hiring around data centers, geopolitical risk, governance and technical implementation. Roles such as chiefs of staff, technical analysts and even newer titles like narrative engineer increasingly sit between technology and the institutions trying to understand or deploy it.
As AI becomes embedded in organizations, the valuable human skill is not simply knowing how to prompt a model. It is knowing when the model should be trusted, what problem it should solve, how its output fits into a larger system and where human judgment still matters.
AI is already useful for structuring information, testing ideas, building software and accelerating defined workflows. Discernment is harder. That distinction will become one of the more important professional advantages over the next few years: knowing what to delegate to the machine and, perhaps more importantly, what not to.
There is an irony here too. Companies such as Mercor are hiring skilled professionals to help train models in increasingly specialized domains. In some cases, the people improving the systems are effectively teaching them portions of the work humans currently perform.
One Rumor to Watch
One item on the radar comes with a very large unconfirmed label.
There is chatter around a possible shift in OpenAI’s relationship with Microsoft and potential deeper business involvement with AWS or Amazon. Nothing discussed in the episode has been officially announced, so treat this as something to watch rather than something that has happened.
If something does emerge, the interesting question will be less about the headline partnership than the economics underneath it: compute costs, infrastructure access and how much leverage frontier AI companies want any single cloud provider to hold over them.
Keep your ears to the ground.
And Finally: Folklore
Not everything interesting this summer involves export controls or autonomous cyberattacks.
Tiktok failed to load.Enable 3rd party cookies or use another browser
Also on my radar is Alex Chidley Uttley, a folklorist and writer specializing in British folklore and the stories embedded in places, traditions and historical figures.
There seems to be a broader cultural return underway toward mythology, folklore and older narratives. Case in point- the extraordinary interest surrounding The Odyssey is one obvious example (upcoming episode on that with an interview soon!). Perhaps that is not unrelated to everything above. When the technological environment changes quickly, people tend to become interested again in the stories that explain where they came from and how people before them made sense of uncertainty.
Sometimes looking forward makes looking backward more interesting.
That’s it for this dispatch.
Thanks for reading Potentia! Join 3.3k+ followers! We genuinely appreciate every like save and follow : ) Your support means the world!
Add to Your Queue!
AI & Technology
OpenAI and Hugging Face Partner to Address Security Incident During Model Evaluation - OpenAI’s account of the autonomous-agent cyber incident and what it suggests about frontier model capabilities.
Security Incident Disclosure — July 2026 - Hugging Face’s side of the same incident.
Modeling an AI Jobs Transition - OpenAI on how AI could reshape work and labor markets.
What Is an AI Trainer? - Mercor on the emerging workforce being paid to train and evaluate AI systems.
OpenAI and Amazon Announce Strategic Partnership - OpenAI’s expanding AWS relationship, including infrastructure, Trainium compute and agent deployment.
Geopolitics & Tech Policy
USTR Takes Action in Forced Labor Section 301 Investigations - The July Section 301 action covering 60 economies.
Trump Administration Drafting Ban on Chinese Data Center Devices - Reuters on optical transceivers becoming the next front in U.S.-China AI infrastructure competition.
FCC Curbs on Chinese Tech Imports Aim to Spur U.S. Production - Brendan Carr on robots, power inverters and the national-security logic behind the expanding restrictions.
AI Infrastructure & Work
New York’s Hyperscale Data Center Moratorium - The state’s one-year pause and emerging framework around energy, water, infrastructure and community costs.
Preparing the Workforce for AI - OpenAI Chief Economist Ronnie Chatterji and Deputy Secretary of Labor Keith Sonderling on reskilling, AI literacy and workforce transition.
Folklore, History & Culture
Alix Chidley-Uttley - Folklorist and writer exploring British landscapes, local history, myths and traditions.
Alix Chidley-Uttley YouTube - Short-form explorations of folklore, wildlife and the stories attached to Britain’s landscape.
The Shropshire Witches Podcast - Alix Chidley-Uttley and Amy Boucher on the hidden, haunted and half-forgotten corners of history and folklore.
The Odyssey - Homer, translated by Emily Wilson; a particularly readable modern version to pair with the current resurgence of interest in the epic.
Watch
New York Announces Its Data Center Moratorium - Governor Kathy Hochul’s announcement.
Preparing the Workforce for AI - OpenAI’s discussion on what AI adoption means for American workers.
